AI RESEARCH
Coherence Maximization Improves Pluralistic Alignment
arXiv CS.CL
•
ArXi:2606.03110v1 Announce Type: new Aligning AI systems with diverse human values requires value specifications grounded in concrete examples, but generating such examples without extensive human supervision remains an open challenge. We investigate what makes these examples effective, using Internal Coherence Maximization (ICM) -- which infers labels by maximizing their mutual predictability -- to generate persona-specific examples that steer a model toward a target group's values, without human supervision.